Additional file 2 of Molecular hallmarks of excitatory and inhibitory neuronal resilience to Alzheimer’s disease

Supplementary Table: Table S1: Differentially expressed genes in AD compared to cognitive resilience in bulk DLPFC tissue. Differential gene expression analysis was performed between AD and resilient subjects (ADvsRES) using ROSMAP bulk RNAseq data. Significant differential expression was determined using adjusted p-value (FDR) < 0.1 (highlighted in red) and |FC| >1.1 criteria. P-values were determined using a linear regression model accounting for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. The column “Direction” shows significant upregulation (“UP”) or downregulation (“DOWN”) in AD versus resilience as defined above. The column “Unique ADvsRES” reports genes that were differentially expressed in ADvsRES and not in ADvsCTRL or RESvsCTRL. The column “Cognitive loss” reports genes also associated with loss of cognition (related with Table S5). Table S2: Differentially expressed genes in AD compared to cognitively healthy controls in bulk DLPFC tissue. Differential gene expression analysis was performed between AD and control subjects (ADvsCTRL) using ROSMAP bulk RNAseq data. Significant differential expression was determined using adjusted p-value (FDR) < 0.1 (highlighted in red) and |FC| >1.1 criteria. P-values were determined using a linear regression model accounting for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. The column “Direction” shows significant upregulation (“UP”) or downregulation (“DOWN”) in AD versus controls as defined above. Table S3: Differentially expressed genes in resilient subjects compared to cognitively healthy controls in bulk DLPFC tissue. Differential gene expression analysis was performed between resilient and control subjects (RESvsCTRL) using ROSMAP bulk RNAseq data. Significant differential expression was determined using adjusted p-value (FDR) < 0.1 (highlighted in red) and |FC| >1.1 criteria. P-values were determined using a linear regression model accounting for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. The column “Direction” shows significant upregulation (“UP”) or downregulation (“DOWN”) in resilience versus controls as defined above. Table S4: Differentially expressed genes in AD compared to subjects classified as presymptomatic in bulk DLPFC tissue. Differential gene expression analysis was performed between AD and presymptomatic subjects (ADvsPRE) using ROSMAP bulk RNAseq data. Significant differential expression was determined using adjusted p-value (FDR) < 0.1 (highlighted in red) and |FC| >1.1 criteria. P-values were determined using a linear regression model accounting for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. The column “Direction” shows significant upregulation (“UP”) or downregulation (“DOWN”) in AD versus presymptomatic as defined above. Table S5: Genes associated with loss of cognition from bulk RNAseq data. The association of gene expression profiles with loss of cognition was analyzed using a proportional odds model (POM) for ordinal categorical data analysis applied to ROSMAP bulk RNAseq data. Gene expression levels were adjusted for confounding covariates prior to performing POM. The POM analysis was applied to the subjects with either no cognitive impairment, mild cognitive impairment, or AD dementia. The POM analyzed cognitive impairment as a function of gene expression and pathology status (plaque and tangle stages). P-values, adjusted for multiple hypothesis comparison using the Bonferroni correction method (column “adj.P.Val”), represent the significance of the association of each gene with cognitive status. The column “Direction” denotes whether the expression of a gene positively or negatively correlates with loss of cognition. The column “ADvsRES” denotes genes that were also differentially expressed between AD and resilient subjects, “ADvsRES dir” reports the direction in AD compared to the resilient group (related with Table S1), and “Change consistency” reports the consistency between the two analyses. Table S6: Dysregulated pathways in AD compared to cognitive resilience in bulk DLPFC. Differential pathway activity analysis was performed between AD and resilient subjects based on ROSMAP bulk RNAseq data. Significant pathway dysregulation was determined using Storey-adjusted p-value (q-value) < 0.1. P-values were determined using a linear regression model that accounts for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. Table S7: Unsupervised clusters of dysregulated pathways in AD compared to cognitive resilience from bulk DLPFC. Cluster membership for each of the 99 dysregulated pathways in ADvsRES (related to Table S6) following unsupervised clustering. Modules of dysregulated pathways in ADvsRES were determined by mapping to the pathway co-expression network followed by the Label Propagation clustering algorithm. The column “cluster” represents the unsupervised clusters assigned by the described analysis. The analysis was performed using the PanomiR package with default parameters. Table S8: Dysregulated pathways in AD compared to cognitively healthy controls in bulk DLPFC. Differential pathway activity analysis was performed between AD and control subjects based on ROSMAP bulk RNAseq data. Significant pathway dysregulation was determined using Storey-adjusted p-value (q-value) < 0.1. p-values were determined using a linear regression model accounting for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. Table S9: Results for the pathway activity analysis in resilient individuals compared to subjects classified as presymptomatic in bulk DLPFC. Differential pathway activity analysis was performed between RES and presymptomatic subjects (RESvs PRE) based on ROSMAP bulk RNAseq data. Significant pathway dysregulation was determined using Storey-adjusted p-value (q-value) < 0.1. P-values were determined using a linear regression model that accounts for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. Table S10: Results for the pathway activity analysis in AD compared to presymptomatic subjects in bulk DLPFC tissue. Differential pathway activity analysis was performed between AD and presymptomatic subjects (ADvs PRE) based on ROSMAP bulk RNAseq data. Significant pathway dysregulation was determined using Storey-adjusted p-value (q-value) < 0.1. P-values were determined using a linear regression model that accounts for confounding covariates such as sequencing batch, RNA integrity number, and postmortem interval. Table S11: Number of subjects and number of cells per group defined in this study for snRNAseq data. Table S12: Distributions for differentially expressed genes per comparison, identified from SnRNAseq data for each major cell type in each brain region investigated. Differential gene expression analyses were performed per major cell type for each brain region by implementing a statistical model by group using the MAST statistical framework in Seurat, after removing non-variable genes. P-values were adjusted using the Bonferroni correction, as recommended for the R package Seurat. Significant differential expression was determined using adjusted p-value (adj-P) < 0.1 and log2FC > 0.2 or log2FC < -0.2 (|FC| >1.1) criteria. The number of downregulated differentially expressed genes (DEGs) in the first group compared to the second group are highlighted in red, and the number of upregulated DEGs in the first group compared to the second group are highlighted in blue. Table S13: Selected differentially expressed genes for major cell types. Differential expression summary results for each brain region across the three comparisons (ADvsRES, ADvsCTRL, and RESvsCTRL) for genes discussed in the text. Differential gene expression analyses were performed per major cell type for each brain region by implementing a statistical model by group using the MAST statistical framework in Seurat, after removing non-variable genes. P-values were adjusted using the Bonferroni correction (adj-P), as recommended for the R package Seurat. Significant differential expression was determined using adjusted adj-p < 0.1 and log2FC > 0.2 or log2FC < -0.2 (|FC| >1.1) criteria. NS: not significant. Direction “NONE” refers to log2FC below our threshold. Extended results tables are available on Synapse (Synapse ID syn63686123). Table S14. Cell annotations for cell subtypes for each brain region. Cluster annotations were generated using the web tool MapMyCells using the Hierarchical algorithm. Table S15: GO enrichment results for the single PPI module of upregulated genes in resilient DLPFC excitatory neurons compared to AD. Performed with Metascape, considering the following databases: STRING, BioGrid, OmniPath, and InWeb_IM. Table S16. GO enrichment results for the single PPI module of upregulated genes in resilient DLPFC excitatory neurons compared to controls. Performed with Metascape, considering the following databases: STRING, BioGrid, OmniPath, and InWeb_IM. Table S17: Results for cell proportion analysis per major cell types for each brain region investigated. Changes in cell composition between groups for each cell subpopulation (subclusters) were detected using a Dirichlet multinomial regression model, while accounting for the proportions of all of the other cell subsets within each major cell type. P-values were adjusted for multiple comparisons using FDR correction (adj-P). Green: adj-P < 0.05, yellow: adj-P < 0.01, red: adj-p < 0.001. Table S18: Phenotypic and clinical characteristics of the BIDMC cohort. Human brain tissue used for immunostaining experiments was collected at Beth Israel Deaconess Medical Center (BIDMC) upon autopsy. Samples were tested for multiple pathologies, including TDP-43. Donors presenting comorbidities, including diabetes, were excluded. Table S19: Ligand-receptor results from cellchat for the SST signaling pathway in the EC. Source cell subtypes included EC:Inh3 and EC:Inh9 in control and resilient subjects, but in AD only EC:Inh3 and EC:Inh11 were identified as sources for the SST pathway in AD. The receptor SSTR1 was only observed in resilience. Table S20: Selected differentially expressed genes for cell subtypes. Differential expression summary results for each brain region across the three comparisons (ADvsRES, ADvsCTRL, and RESvsCTRL) for genes discussed in the text. Differential gene expression analyses were performed per cell subtype for each brain region by implementing a statistical model by group using the MAST statistical framework in Seurat, after removing non-variable genes. P-values were adjusted using the Bonferroni correction (adj-P), as recommended for the R package Seurat. Significant differential expression was determined using adjusted adj-p < 0.1 and log2FC > 0.2 or log2FC < -0.2 (|FC| >1.1) criteria. NS: not significant. Direction “NONE” refers to log2FC below our threshold. Table S21: Interaction partners of RBFOX1 identified as rare variant-associated genes and marker genes for DLPFC:Inh1 neurons. Table S22: Functional enrichment of RBFOX1 partners, also identified as rare variant associated genes and marker genes for DLPFC:Inh1 neurons.